Published network models out there have 2D (or higher-dimensional) spatial topology incorporating distance-dependent connection probabilities and connection delays, etc.
The question is: should we aim to add support for computing filter kernels applicable to such networks?
This issue should summarize what is needed.